Machine learning in ophthalmology: A comprehensive review on glaucoma detection and diagnosis

Abstract

Glaucoma is a leading cause of irreversible blindness worldwide, and its early detection remains a critical challenge in ophthalmic care. Recent advances in machine learning (ML), particularly deep learning techniques such as convolutional neural networks (CNNs), have shown immense promise in improving diagnostic accuracy, efficiency, and accessibility in glaucoma screening. This paper provides a comprehensive review of 20 recent studies (2020-2025) that apply ML models to glaucoma diagnosis using diverse data sources, including fundus photographs, optical coherence tomography (OCT), and visual field test results. The findings reveal that ensemble models, multimodal learning, and explainable AI significantly enhance predictive performance and clinical trust. Additionally, portable AI solutions demonstrate potential for addressing disparities in low-resource settings. Despite these advancements, the paper identifies key research gaps, such as limited generalizability due to homogenous datasets, lack of real-time clinical integration, and insufficient focus on algorithm interpretability and ethical deployment. The review concludes with future suggestions including the development of diverse datasets, integration of multimodal and longitudinal data, adoption of transparent AI frameworks, and emphasis on interdisciplinary collaboration. These directions are essential for translating ML-based glaucoma detection systems from experimental models into reliable tools for routine ophthalmic care.